myrtle
I can provide a concrete example: `run.sh` in https://ds0.me/klayout-drc-perf.tar.xz - this is about a 2.9mm x 3.5mm SoC on sky130 built using Coriolis `"space" in: sky130A_mr.drc:332` takes 2465s it then...
I will investigate, but unfortunately this is a bit of an ongoing issue in nextpnr in general with routing congestion - particularly nextpnr-nexus - so I can't promise a quick...
Some of the router1 slowness is probably around Vcc routing on the nexus which is a specific bug I'll look at once I've fixed the DSP placement one. I'm also...
https://github.com/gatecat/prjoxide/pull/15 and https://github.com/YosysHQ/nextpnr/pull/822 gets the `router1` runtime down to about 40min. I'm looking at some further tweaks for router1 performance though, as this has never been optimised for nexus before.
There might be scope for more fine-tuning, but https://github.com/YosysHQ/nextpnr/pull/823 on top of those gets `router1` down to 20min. It might now be possible to do some comparison between router1 and...
Still working on an actually functionally correct implementation but it looks like LUT permutation should help a lot, although we might have to switch nexus back to router1 for a...
I need to sort out the existing patches before I make it into a PR, and also fix a separate issue that has been triggered in router2 (this also sets...
lutperm has now been merged; however I'd recommend keeping `--router router1` on this project for now for more reliable timing and convergence (at the expense of increased runtime).
Unfortunately the problem here is an abc9 bug, I think, which I have seen on occasion for the Yosys Cyclone V support; where abc9 touches a carry chain that it's...